You've wondered it too. You get a decent answer from ChatGPT and a little voice asks, would Claude have done this cleaner? Would Gemini catch the edge case?
The honest answer is you usually don't check. Because checking means setting the whole thing up again somewhere else.
Here's the trap. A real comparison only counts if both models get the same context. Same prompt, same constraints, same history. The second you retype your setup into a new tab, you've changed the input. You shortened something, reworded something, dropped the detail you forgot. Now you're not comparing models. You're comparing two slightly different conversations and pretending they're the same.
Retyping the "identical" prompt for the third time and calling it science.
So I mostly didn't bother. I'd take whatever the first model gave me and move on, with that small itch of not knowing.
What changed is dumb and I almost feel silly about it. I started moving the exact conversation between models instead of rebuilding it.
There's a Chrome extension, Migrate My AI Chat, that puts a little button on the AI sites you already use. I run a full thread in ChatGPT until I've got a real answer. Then I send that same conversation to Claude. Same history, byte for byte. Claude picks it up and answers the same situation. Then I do it again with Gemini.
Now I'm actually comparing the models. Not my memory of the prompt.
Me, popcorn in hand, watching three models answer the exact same thread.
You learn things fast this way. One model loves to refactor your whole file when you asked for a two-line fix. One hedges. One catches the null case the others walked right past. You stop arguing about which AI is "best" in the abstract, because best depends entirely on what you're doing, and now you can see it on your own work in about a minute.
Finally judging the models on the same input instead of vibes. Chef's kiss.
It helps that nothing leaves the browser. No account, no server in the middle. The conversation just gets handed from one tab to the next. For comparing answers about actual code I'm working on, I wasn't going to use anything that phoned home.
The small thing that still makes me grin: the receiving model doesn't act confused. It doesn't say "I don't have the earlier context." It just answers, like it had been reading along the entire time.
If you've ever wondered whether the other model would've done it better, you can finally just look. It's free.
Top comments (0)